AI Model Unlocks Drug Safety Insights With World’s Largest Adverse Event Database

A Cedars-Sinai–led study introduces OnSIDES, the most comprehensive AI-powered database of adverse drug events ever assembled, offering a transformative tool for preventing medication-related deaths and reshaping drug research worldwide.

Research: OnSIDES database: Extracting adverse drug events from drug labels using natural language processing models. Image Credit: 3rdtimeluckystudio / Shutterstock

A multicenter study led by Cedars-Sinai created a database of adverse medication events, the fourth leading cause of death in the United States and a medical issue costing more than $500 billion annually.

The findings, published in the peer-reviewed journal Med, demonstrate how AI can improve drug safety, support drug discovery, and enhance understanding of medication risks.

The database is called OnSIDES (ON-label SIDE effectS resource). It is free and publicly available on GitHub.

"OnSIDES provides the most comprehensive and up-to-date database of adverse drug events from drug labels," said Nicholas Tatonetti, PhD, vice chair of Computational Biomedicine at Cedars-Sinai and corresponding study author. "This work enables researchers and clinicians to systematically study drug safety." 

Adverse drug events are unintended, harmful events related to the usage of medication and are the fifth leading cause of death internationally. Experts believe half of all adverse drug events are preventable.

"While many drug safety studies are conducted on individual medications during clinical trials and through post-marketing surveillance programs, far fewer studies have studied the occurrence of adverse drug events more broadly," said Tatonetti, also the associate director for Computational Oncology at Cedars-Sinai Cancer. "The lack of broadscale studies may be attributed in part to the array of medications and the complexity of drug interactions, as well as the lack of standardized data publicly available."

The OnSIDES model analyzed 3,233 unique drug ingredient combinations extracted from 47,211 labels and identified over 3.6 million pairs of medications and adverse drug events. This work was also expanded to labels from countries outside of the U.S., revealing differences in how adverse drug events are reported internationally.

By using artificial intelligence to extract adverse drug events from drug labels, investigators improved access to structured, machine-readable data. This ultimately made it easier to identify drug risks, predict new drug uses, and enhance patient safety.

"This resource supports drug repurposing, pharmacovigilance, and AI-driven drug discovery," said Jason Moore, PhD, chair of the Department of Computational Biomedicine at Cedars-Sinai. "We are hopeful that future research can build on OnSIDES to develop better predictive models, personalized medicine approaches, and regulatory insights, ultimately leading to safer medications and more informed clinical decision-making worldwide."